Low-Power Error Correction for Mobile Storage



Similar documents
The Efficient LDPC DSP System for SSD

NAND Flash Architecture and Specification Trends

Important Differences Between Consumer and Enterprise Flash Architectures

NAND Flash Architecture and Specification Trends

3D NAND Technology Implications to Enterprise Storage Applications

Solid State Drive Technology

In-Block Level Redundancy Management for Flash Storage System

The Evolving NAND Flash Business Model for SSD. Steffen Hellmold VP BD, SandForce

Data Retention in MLC NAND Flash Memory: Characterization, Optimization, and Recovery

Data Storage Time Sensitive ECC Schemes for MLC NAND Flash Memories

SLC vs MLC: Proper Flash Selection for SSDs in Industrial, Military and Avionic Applications. A TCS Space & Component Technology White Paper

Design of a High-speed and large-capacity NAND Flash storage system based on Fiber Acquisition

SLC vs MLC: Which is best for high-reliability apps?

Data Distribution Algorithms for Reliable. Reliable Parallel Storage on Flash Memories

Improved NAND Flash Memories Storage Reliablity Using Nonlinear Multi Error Correction Codes

On the Use of Strong BCH Codes for Improving Multilevel NAND Flash Memory Storage Capacity

Flash & DRAM Si Scaling Challenges, Emerging Non-Volatile Memory Technology Enablement - Implications to Enterprise Storage and Server Compute systems

BER-Based Wear Leveling and Bad Block Management for NAND Flash

Improved NAND Flash Memories Storage Reliablity Using Nonlinear Multi Error Correction Codes

A Comparison of Client and Enterprise SSD Data Path Protection

Programming Matters. MLC NAND Reliability and Best Practices for Data Retention. Data I/O Corporation. Anthony Ambrose President & CEO

A Storage Architecture for High Speed Signal Processing: Embedding RAID 0 on FPGA

What Types of ECC Should Be Used on Flash Memory?

Flash Technology Update from Micron and Intel

Disks and RAID. Profs. Bracy and Van Renesse. based on slides by Prof. Sirer

Industrial Flash Storage Module

Power Reduction Techniques in the SoC Clock Network. Clock Power

SSD Server Hard Drives for IBM

Make A Right Choice -NAND Flash As Cache And Beyond

Definition of RAID Levels

Technologies Supporting Evolution of SSDs

HP Smart Array Controllers and basic RAID performance factors

Disk Storage & Dependability

Temperature Considerations for Industrial Embedded SSDs

Data Retention in MLC NAND Flash Memory: Characterization, Optimization, and Recovery

Don t Let RAID Raid the Lifetime of Your SSD Array

enabling Ultra-High Bandwidth Scalable SSDs with HLnand

Scaling from Datacenter to Client

End to End Data Path Protection

NAND Flash FAQ. Eureka Technology. apn5_87. NAND Flash FAQ

Flash for Databases. September 22, 2015 Peter Zaitsev Percona

Managing the evolution of Flash : beyond memory to storage

Intel Solid-State Drive 320 Series

SOS: Software-Based Out-of-Order Scheduling for High-Performance NAND Flash-Based SSDs

Flash Memory. Jian-Jia Chen (Slides are based on Yuan-Hao Chang) TU Dortmund Informatik 12 Germany 2015 年 01 月 27 日. technische universität dortmund

Samsung 3bit 3D V-NAND technology

Chapter Introduction. Storage and Other I/O Topics. p. 570( 頁 585) Fig I/O devices can be characterized by. I/O bus connections

Understanding endurance and performance characteristics of HP solid state drives

Programming NAND devices

Flash Controller Architecture for All Flash Arrays

RAID. RAID 0 No redundancy ( AID?) Just stripe data over multiple disks But it does improve performance. Chapter 6 Storage and Other I/O Topics 29

Flash 101. Violin Memory Switzerland. Violin Memory Inc. Proprietary 1

Yaffs NAND Flash Failure Mitigation

Booting from NAND Flash Memory

Charge-Trapping (CT) Flash and 3D NAND Flash Hang-Ting Lue

760 Veterans Circle, Warminster, PA Technical Proposal. Submitted by: ACT/Technico 760 Veterans Circle Warminster, PA

Samsung 2bit 3D V-NAND technology

Trends in NAND Flash Memory Error Correction

White Paper. Avoiding premature failure of NAND Flash memory. Inhalt

Cloud Storage. Parallels. Performance Benchmark Results. White Paper.

Solid State Drives Data Reliability and Lifetime. Abstract

SRAM Scaling Limit: Its Circuit & Architecture Solutions

SSDs and RAID: What s the right strategy. Paul Goodwin VP Product Development Avant Technology

7a. System-on-chip design and prototyping platforms

Maximizing Your Server Memory and Storage Investments with Windows Server 2012 R2

Price/performance Modern Memory Hierarchy

Sentinel-2 MMFU The first European Mass Memory System based on NAND-Flash Storage Technology

Design and Implementation of a Storage Repository Using Commonality Factoring. IEEE/NASA MSST2003 April 7-10, 2003 Eric W. Olsen

AN1837. Non-Volatile Memory Technology Overview By Stephen Ledford Non-Volatile Memory Technology Center Austin, Texas.

Comparison of NAND Flash Technologies Used in Solid- State Storage

International Journal of Electronics and Computer Science Engineering 1482

A Close Look at PCI Express SSDs. Shirish Jamthe Director of System Engineering Virident Systems, Inc. August 2011

Nonlinear Multi-Error Correction Codes for Reliable NAND Flash Memories

An Overview of Flash Storage for Databases

IOmark- VDI. Nimbus Data Gemini Test Report: VDI a Test Report Date: 6, September

Choosing the Right NAND Flash Memory Technology

Scalus Winter School Storage Systems

A case study of mobile SoC architecture design based on transaction-level modeling

A 10,000 Frames/s 0.18 µm CMOS Digital Pixel Sensor with Pixel-Level Memory

Filesystems Performance in GNU/Linux Multi-Disk Data Storage

Database!Fatal!Flash!Flaws!No!One! Talks!About!!!

COS 318: Operating Systems. Storage Devices. Kai Li Computer Science Department Princeton University. (

RAID Performance Analysis

Indexing on Solid State Drives based on Flash Memory

Database Hardware Selection Guidelines

Solid State Technology What s New?

Linux flash file systems JFFS2 vs UBIFS

Flash-optimized Data Progression

Implementation of Full -Parallelism AES Encryption and Decryption

Accelerating Server Storage Performance on Lenovo ThinkServer

The Shortcut Guide to Balancing Storage Costs and Performance with Hybrid Storage

Design of a Wireless Medical Monitoring System * Chavabathina Lavanya 1 G.Manikumar 2

FUNDAMENTALS OF MODERN SPECTRAL ANALYSIS. Matthew T. Hunter, Ph.D.

Solid State Storage in Massive Data Environments Erik Eyberg

Security & Chip Card ICs SLE 44R35S / Mifare

SLC vs. MLC: An Analysis of Flash Memory

Transcription:

Low-Power Error Correction for Mobile Storage Jeff Yang Principle Engineer Silicon Motion 1

Power Consumption The ECC engine will consume a great percentage of power in the controller Both RAID and LDPC become crucial to high reliable flash controller, which is employed in embedded applications. Stronger correction capability Longer ECC chunk (e.g., 2KB BCH is stronger than 1KB, and 2KB LDPC is much stronger than 1KB) Higher decoder resolution and parallelism architecture will reduce the decoding latency. Higher power consumption. Wider range corrupt data recovery Larger RAID recovery range will provide wider range data recovery. Consume more memory size and calculation logic. Higher power consumption. In normal operation For RAID engine, only the encoder part is active all the time. LDPC uses the hard-decoding only, but the encoder is always active in each flash write action. Power consumption reduction in hot region RAID encoding LDPC decoding SMI Confidential 2

RAID Engine The SSD module failure rate will be the Key for TLC adoption on SSD. Latest Planar NAND might be 1znm. Bit column, sense-amp, address decoder, or charge pump corrupted WL lining or WL to WL short. First popular 3D structure on BiCs. Inherit most of failure cases from 2D. High aspect ratio etching and larger number of layers control will contribute more failure cases. No Erase/Program failure, but with reading ECC failure. Need RAID engine to do the error recovery. In 3D TLC, three WL lining may cause data loss over 9 pages. High efficiency RAID with erasure decoding. Encoding with high efficiency memory access. Memory access dominates the encoding power. Low RAID recovery trigger rate. [Ref]: VLSI symp. 2009, pp.136-137, VLSI symp. 2008, pp. 192-193 VLSI symp. 2010, pp. 131-132, IEEE Trans. Electron Devices, vol.58, no.5, pp. 1006-1014, Apr.2011 Lower grade TLC flash with RAID to have 2.6K endurance 3

DSP Engine Power Consumption NAND Flash characteristics on Endurance and Retention. Retention capability on TLC degrades rapidly as P/E cycle increases. Reduce the power with Retention-aware algorithm: Predict the sensing point in the first read. Joint the soft-decoding and the Vt-tracking. Minimize the number of read operation The error bit number is the key factor of power consumption. Sensing the data which has lowest bit error rate Predict error types and select the decoding strategy. 4

NAND characteristic. SLC/MLC/TLC 5

TLC Error Behavior on Endurance X: Number of error bit Y: Number of ECC chunk Endurance 2K, 90% of the error bit number is less than 30 bits 2K: CSB-RBER = 2.7e-3 1K: CSB-RBER = 1.3e-3 Sensing the data with lowest error bit. Predict the error case and select the Soft-decoding/RAID strategy. trigger rate X: P/E cycle After the 2.7K cycling, it needs LDPC soft-decoding. RAID redundant overhead ~ 0.07%. Double strong endurance from LDPC hard to soft decoding. First RAID recovery occurs on 4.5K cycles. 6

LDPC Low Power Decoding Technology A mw 2A mw 3A mw Quantization Optimization for the Harddecoding mode. Change floating point calculation system 3A mw Traditional LDPC engine structure is here Clock gated control for shift register Smart Error prediction Retiming optimization Adaptive clipping Future work New decoding algo? 1.3A mw Power efficient LDPC engine structure 7

1KB BCH vs. LDPC decoding power consumption TSMC40nm on TLC flash Error bit range /1KB BCH LDPC NOTE 1~20 bit A mw 0.9A mw ~400MB/sec 20~40 bit B mw 1.4B mw ~400MB/sec 40~60 bit C mw 2.6 C mw ~300MB/sec Endurance BCH LDPC NOTE 0~1K P/E A mw 0.9A mw Hard-bit only 1.~2K B mw 1.4B mw Hard-bit only 2K~5K NA 2.6C mw, poor throughput soft-decoding trigger frequently. If requirement is 1K cycles, Will be the BCH enough to support the TLC? Flash Memory Summit 2013 8

After Data Retention at 500P/E Flash Memory Summit 2013 9

Comparison between 1KB and 2KB Error bit under 2KB chunk is 110bit. Almost 100% correctable rate for Hard-decision only. 851 MB/sec under 400MHz operation frequency. When the error bit under 1KB chunk is 60bit. Almost 100% correctable rate for Hard-decision only. 330 MB/sec under 400MHz operation frequency. 1KB vs. 2KB under the same error condition 1KB/60bit 2KB/110bit Area (gate) 1A 1.8A TP (@400MHz) 330 MB/sec 851 MB/sec Power (W) 1A 2.15A Power Efficiency 1A MB/J 1.2A MB/J Area Efficiency (per gate) 1A B/s 1.4A B /s UBER=1e-15 RBER=1e-2 RBER=1.25e-2 10 SMI Confidential

Ideal Solution for TLC-based SSD LDPC 1K+124B, t=~120 RBER= 1e-2, UBER=1e-15 BCH 2KB+228B, t=121bit, RBER=3.5e-2, UBER=1e-15 P/E = 2379 BCH 1KB+124B, t=70bit, RBER=3.1e-2, UBER=1e-15 P/E = 2164 The 2KB BCH engine only has limited improvement on endurance and retention. LDPC ideally fits TLC and maximizes the endurance of TLC-based SSD solution. LDPC + RAID is the ONLY reliable solution for TLC-based SSD. 11

Q & A Disclaimer Notice Although efforts were made to verify the completeness and accuracy of the information contained in this presentation, it is provided as is as of the date of this document and always subject to change. 12